Robinson Meyer is a staff writer at The Atlantic, where he covers climate change and technology.
Descartes Labs lets you point-and-hop between features in China and the United States.
At this moment in history, there are more satellites photographing Earth from orbit than just about anyone knows what to do with. Planet, Inc., has more than 150 orbiting cameras, each the size of a shoebox. DigitalGlobe has five dump-truck-sized sensors. And more startups are planning to launch their own.
What should we do with all that imagery? How can we search it and process it? Descartes Labs, a startup that uses machine learning to identify crop health and other economic indicators in satellite imagery, has created a tool to better index and surf through it. They call it Geovisual Search.
Geovisual Search allows users to find similar-looking objects in aerial maps of China, the United States, and the world. It’s free and available online right now. Click on a visible feature—like an oil tank, an empty swimming pool, or a stack of shipping containers—and Geovisual Search will find other objects like it on the map.
Here’s a search, for instance, for solar farm-looking features in China:
“Imagine these big data sets coming along from Planet. Suddenly you’re getting daily pictures of the globe. You kind of want to count these things, every single day, and watch how they change through time,” says Mark Johnson, the CEO of Descartes Labs.
“The neural nets that we trained here are the beginning of counting oil tanks, or buildings, or windmills. Imagine we wanted to look at sustainable energy infrastructure—solar farms, solar panels on roof—you could start to think about counting their growth through time. You start to get really interesting data streams,” he told me.
It’s a legitimately cool way to search satellite imagery, and it’s great to be able to surf through the terrain of China and the United States as a whole. It reminded me of Terrapattern, an art project created by artists and geographers at Carnegie Mellon University last summer. Terrapattern had a near-identical interface and near-identical capabilities to Descartes Search, but it only accessed certain urban areas in the U.S., including Pittsburgh, New York, and the Bay Area.
The Decartes team tips their hat to Terrapattern in their announcement blogpost, calling the earlier project a “ground-breaking demonstration of visual search over satellite imagery.”
“We loved it. The demo aligned with many ideas we had been kicking around at Descartes Labs, and it was great to see somebody just go out and do it,” the blog post says.
Despite this admiration, Descartes only ran their implementation past the Terrapattern team 12 hours before its release. “Their approach is virtually the same as what we did a year ago, with some tweaks to deal with scale,” said Golan Levin, who led the Carnegie Mellon team, in an email.
“It’s quite typical for new-media artworks to, er, ‘inspire’ commercial projects—this is unfortunately quite common,” he said. “Since our team is artists and students and academics, the chance or option to have collaborated would have been much more fun.”
In fact, Levin has written about how Google Streetview, Sony Eyetoy, and a Nike product called the “Chalkbot” were all inspired by new-media artistic experiments. He added that Terrapattern is now working with a major satellite-imagery provider and a design firm to create a similarly scaled-up version of its product.
Perhaps this method of searching a geographic environment will eventually have the same renown as Google Streetview. If the sheer amount of new daily satellite imagery continues to expand, it seems like a possible fate. For its part, Descartes plans to keep expanding the use of machine-learning algorithms on satellite imagery. It will also continue producing its corn-health forecasts.
This post originally appeared on The Atlantic.